The Hidden Rivers of Sinking Air That Wrap Every Thunderstorm

Every thunderstorm carries with it an invisible companion. Around the edges of nearly every convective cloud, a thin curtain of air slides downward, hugging the cloud like a descending shell. Scientists call this feature a subsiding shell, and although it has been studied for decades in small, fair-weather cumulus clouds, its behavior around towering tropical thunderstorms has remained poorly understood. Now, a new study by Quinlan R. Mulhern, John M. Peters, and Jake P. Mulholland, published in Atmospheric Chemistry and Physics, has disentangled the forces that drive these descending curtains in deep convection, and the answer overturns a long-standing assumption about why downdrafts are stronger in drier atmospheres.

The stakes are larger than they might first appear. Subsiding shells are not a curiosity of cloud physics; they are a leading contributor to downdraft mass flux in the atmosphere. Previous work by Glenn and Krueger found that the compensatory downward mass flux associated with deep convective subsiding shells amounts to roughly five to ten percent of the upward flux at any given level. Because global climate models cannot explicitly resolve individual thunderstorms, they rely on cumulus parameterizations to represent these sub-grid exchanges of mass, heat, and moisture. Many of those schemes assume downdrafts simply displace the mean properties of a model grid cell downward, an assumption that the near-cloud environment routinely violates. Getting the shells right, in other words, is a prerequisite for getting the climate right.

To probe the shells, the team ran two large-eddy simulations of deep moist convection using the Cloud Model 1 code, with an isotropic 100-meter grid spacing across a 200-square-kilometer domain extending 20 kilometers high. The simulations were initialized with composite soundings drawn from the Green Ocean Amazon campaign, a major field effort conducted in the Amazon Rainforest in 2014 and 2015. The two soundings represented the Amazonian wet and dry seasons, which share nearly identical temperature profiles but differ sharply in free-tropospheric relative humidity. By holding the temperature profiles equal and nudging the domain-averaged moisture above four kilometers back toward its initial state, the researchers isolated the influence of humidity on convection, creating a clean natural experiment inside the computer.

The simulations reproduced the observed atmosphere remarkably well. Updraft tops reached 12 to 16 kilometers, consistent with radar echo tops observed during the field campaign, and time-averaged maximum vertical velocities were greater in the dry season simulation, at roughly 31 meters per second, than in the wet season simulation, at about 27 meters per second. A simple entraining plume model helped explain why: the drier lower troposphere raised the lifting condensation level, producing wider updrafts with smaller fractional entrainment rates, while steeper low-level lapse rates boosted buoyancy. The stage was set to ask what these stronger updrafts meant for the air sinking around them.

The heart of the study was a Lagrangian trajectory analysis of extraordinary scale. Nearly five and a half million inert parcels were seeded uniformly through the domain, and their positions, vertical velocities, buoyancy, and pressure-gradient accelerations were recorded every 30 seconds during the analysis hour. Parcels qualified as shell parcels if they began descending from at least five kilometers, reached a vertical velocity of minus 1.5 meters per second or stronger, displaced at least 500 meters, and spent at least 90 percent of their downdraft lifetime within a subsiding shell, defined as the continuous ring of negative vertical velocity immediately adjacent to cloud edge.

The results were striking. Shell downdrafts in the dry simulation reached a mean minimum vertical velocity of minus 3.44 meters per second, compared with minus 2.78 meters per second in the wet simulation, making the dry-season shells about 23.5 percent stronger, a difference that held statistical significance through essentially the entire downdraft. This matched radar wind profiler observations from the Amazon, which had recorded more frequent strong downdrafts in the dry season. But the reason for the difference was not what many had assumed.

For years, the leading hypothesis held that drier environmental air enhances evaporative cooling at cloud edge, chilling the air, making it negatively buoyant, and thereby driving stronger descent. The trajectory analysis confirmed that shell parcels are indeed negatively buoyant, with dry-season parcels reaching a minimum mean buoyancy of minus 0.021 meters per second squared versus minus 0.016 in the wet case. Yet when the researchers computed effective buoyancy, which combines thermal buoyancy with the vertical perturbation pressure gradient that the atmosphere generates in response to buoyancy anomalies, the seasonal difference nearly vanished. The stronger negative buoyancy in the dry case was almost entirely offset by a stronger upward-directed buoyancy pressure acceleration, a self-regulating thermodynamic response that canceled the apparent advantage of extra evaporation.

The true driver of the seasonal difference lay elsewhere: in dynamic pressure accelerations tied to the toroidal circulations of rising cloud thermals. As a thermal ascends, it carries a region of low perturbation pressure at its core, embedded within the swirling vortex-like circulation that surrounds it. Air parcels passing down the cloud flank are, in the authors’ words, essentially sling-shotted downward by the descending branch of this circulation. The magnitude of this toroidal low scales with updraft strength and horizontal buoyancy gradients, so the stronger dry-season updrafts produced stronger dynamic pressure minima and correspondingly larger downward accelerations. When the team decomposed the kinetic energy budget of shell parcels, the dry-season advantage in dynamic nonlinear pressure energy emerged as the dominant explanation for the stronger downdrafts, while the effective buoyancy contributions were nearly identical between seasons.

The study also revealed that cloud depth matters profoundly. Splitting the shell population into deep convection, with cloud tops above five kilometers, and congestus convection below that threshold, the researchers found that deep shells reached minimum vertical velocities of minus 3.66 and minus 2.96 meters per second in the dry and wet simulations respectively, while congestus shells were markedly weaker, with dry congestus downdrafts a full meter per second slower than their deep counterparts. In the congestus regime, the dynamic pressure contribution collapsed, and negative effective buoyancy from persistent evaporative cooling became the primary driver, so that the stronger dry-season congestus shells were buoyancy-driven rather than dynamically driven. This regime dependence carries a direct warning for modelers: parameterizations cannot assume a single universal mechanism for shell formation across all cloud types.

The composite structures painted a coherent physical picture of a shell’s life. Near cloud top, parcels are accelerated downward both by the dynamic pressure dipole of the toroidal circulation and by negative buoyancy from evaporating cloud droplets at the edge. As they descend past the pressure minimum, the pressure acceleration reverses to point upward, and once parcels overshoot their level of neutral buoyancy, thermal buoyancy turns positive as well. Descent then ceases, typically after a displacement of only about a kilometer, with turbulent mixing at the updraft-shell interface providing a persistent drag throughout. The shells themselves proved narrower than expected, averaging roughly 480 meters wide for deep convection and 420 meters for congestus, with width depending mainly on cloud depth rather than environmental humidity. The authors caution that acceleration magnitudes remain sensitive to choices of microphysics and turbulence schemes, and they recommend that future work extend these findings to strongly sheared environments and to observational campaigns such as ESCAPE. Ultimately, they argue, the dynamics of subsiding shells deserve explicit treatment in cumulus parameterizations, so that climate models can finally capture the hidden rivers of sinking air that every thunderstorm drags down with it.

Subject of Research: Driving mechanisms of subsiding shells surrounding deep convective clouds in large-eddy simulations

Article Title: Driving mechanisms for subsiding shells in simulations of deep moist convection

Article References: Mulhern, Q. R., Peters, J. M., & Mulholland, J. P. (2026). Driving mechanisms for subsiding shells in simulations of deep moist convection. Atmospheric Chemistry and Physics, 26(19), 13959-13982. https://doi.org/10.5194/acp-26-13959-2026

Image Credits: AI Generated

DOI: 10.5194/acp-26-13959-2026

Keywords: subsiding shells, deep convection, downdrafts, large-eddy simulation, cumulus parameterization, evaporative cooling, perturbation pressure, toroidal circulation, Amazon, GOAMAZON, cloud dynamics, relative humidity